JP2022175849A - Ultra high speed image sensor - Google Patents

Abstract

To provide an image sensor and a photographing device capable of controlling the delay in the horizontal movement of a signal electron in the pixel of the image sensor to make the shooting speed ultra-high, and taking 10 to 40 consecutive images at a shooting speed equivalent to 100 billion images/second, or 100 images continuously at a shooting speed equivalent to 1 billion images/second.SOLUTION: In a sensor chip 136, each pixel is obtained by laminating an electrically insulating layer 143 parallel to a C plane parallel to a plane on which pixels are arranged, a circuit diffusion layer 141 which is a semiconductor layer, and a photoelectric conversion layer 140 which is a layer that generates charges by electromagnetic waves or charged particles in an A direction perpendicular to the plane on which the pixels are arranged from the insulating layer 143, and an electrode layer 144 in which L (L≥1) electrodes 149 are arranged in a B direction opposite to the A direction, and a surface on the photoelectric conversion layer side of adjacent surfaces of the circuit diffusion layer and the photoelectric conversion layer is substantially smaller than a surface on the circuit diffusion layer side, and one of the electrodes 149 includes two or more contacts that electrically join wires carrying different voltages.SELECTED DRAWING: Figure 12

Description

The present invention relates to ultra-high-speed imaging technology, and more particularly to ultra-high-speed image sensors.

(multi-charge collection gate image sensor)
Non-Patent Document 1 discloses an image sensor with a time resolution of 10 nanoseconds (100 million sheets/second).

Non-Patent Document 1 discloses an example in which the image sensor continuously captures the flight of light. This document is the first in the world to disclose continuous shooting of light flight using an image sensor.

FIG. 1 shows 5 images from the 6th image (code 1) to the 10th image out of the 10 consecutive images.

Also shown in FIG. 1 is a single image 2 that was captured by a conventional video camera. A typical video camera takes one image every 1/60 second (16,666,666 nanoseconds) on odd or even rows. Therefore, an image of flying light for a total of 100 nanoseconds for capturing ten continuous images with a time resolution of 10 nanoseconds will overlap as a continuous line in one of the images with a normal video camera. .

FIG. 2 shows a cross-sectional view 3 of one pixel of an image sensor with a time resolution of 10 nanoseconds, a layout diagram 4 of electrodes on the front side (lower side in the figure), and photoelectric conversion with light 6 incident on the back side (upper side in the figure) 5. Example 8 in which the trajectory of electrons generated in the layer (photodiode) 7 was reproduced by Monte Carlo simulation, and the time (horizontal axis) from the position where the electron was generated to a certain depth (vertical axis) and the surface side The time spread 9 on arrival is shown.

There are six charge collection gates 11 and 12 arranged radially from the pixel center 10 on the front surface side. ing. A voltage VB is applied to the back surface, and electrons are led to the front surface side at high speed in a depletion layer generated by a large reverse voltage between VB and VH.

These electrons are collected at the pixel center 10 by the p-well 13, and are transferred to the center 10 of the circuit diffusion layer 14 on the surface side, and further to the charge storage gate 15 connected to the charge collection gate 11 via the charge collection gate 11 applied with VH. lead.

By sequentially applying VH to the electrodes of the charge collection gates and VL to the others at short intervals, the electrons are individually induced and stored in the six charge storage gates in sequence.

As a result, image signals for six consecutive images are stored in the charge storage gates.

This ends the shooting.

After photographing is completed, the image signal is transferred to the readout CCD gate 16 arranged at the outermost edge. Since it is isolated from the charge storage gate 15 by the barrier gate 17, a high voltage is applied to the barrier gate to release the isolation, the voltage of the charge storage gate is lowered, and the signal charge is read out and transferred to the CCD gate.

Further, the image signal is transferred downward on the reading CCD gate 16 and led out of the light receiving surface. In a general CCD image sensor, readout CCDs are arranged vertically and linearly. Since they are arranged in the shape of a letter, they are transferred downward along the read-out CCDs arranged in a vertical line (not shown).

Furthermore, it is transferred to a readout circuit through a horizontal CCD arranged outside the light receiving surface, converted into a voltage signal by floating diffusion, and read out to a circuit outside the image sensor at a low speed, that is, with low noise. The path from the readout CCD to the floating diffusion is not shown because it is a common readout system for CCDs.

Since all the transfer paths to the floating diffusion are composed of embedded CCDs, in principle, data can be transferred to the floating diffusion without noise.

When reading from the floating diffusion, even if it is made by a perfect process, in principle, readout noise proportional to the square root of the readout speed intervenes.

Therefore, if the image sensor is sufficiently cooled to suppress the generation of dark current and low-speed readout is performed, virtually noiseless ultra-high-speed imaging can be realized in practice.

An image sensor having the structure of FIG. 2 is called a "multiple charge collection gate image sensor".

In this way, an image sensor that has multiple image signal storage areas for each pixel in or near each pixel is called an image sensor that expresses its structure and function. It is called a "burst image sensor" as a name.

In a normal image sensor, image signals generated during photographing are converted into voltage signals and continuously transferred to a memory area outside the image sensor for recording.

At this time, the imaging speed of the image sensor is limited by the time required to read out signals for one frame consisting of many pixels via the limited number of signal readout lines of one image sensor. is.

The burst image sensor does not read image signals to the outside during shooting. Consecutive images can be taken at extremely short time intervals by transferring the generated image signals to a storage area in or near each minute pixel of only about 10 microns. For example, in the example of FIG. 2, continuous imaging can be performed at extremely short time intervals during which signal electrons are transferred from the pixel center 10 to the charge storage gate 15 .

The number of continuous shots taken by the multi-charge collection gate image sensor is 6, which is small. This is because the purpose is to prove the principle. The purpose was achieved by photographing the flight of light.

Furthermore, one of the six charge storage gates was connected to a drain wiring (not shown) in order to prevent electrons generated immediately before and after imaging from entering the charge storage section. Therefore, the actual number of continuous shots is five.

Since the number of shots is small, the camera also incorporates a function that alternately shoots pixels on even-numbered rows and pixels on odd-numbered rows. Therefore, even 10 shots can be taken. However, the sensitivity will be 1/2.

Furthermore, if 2×2 pixels are defined as one macro pixel and four element pixels in the macro pixel are sequentially photographed, the number of photographed images becomes 20. However, the sensitivity becomes 1/4. However, since the image sensor is a proof-of-principle sensor, it does not incorporate the function of quadrupling the number of shots.

Even without using a multi-charge collection gate structure, an image sensor with a macro pixel consisting of J×J element pixels can capture J ² images in a very short time interval when the element pixels are sequentially captured.

However, the sensitivity becomes ¹ /J2.

Since the amount of light incident on one image is small in ultra-high-speed photography, a drop in sensitivity is critical and is rarely used.

A major feature of the multi-charge collection gate structure is that, in principle, the aperture ratio is 100%, and five images can be taken at an ultra-high speed with virtually no noise without a decrease in sensitivity.

If J is suppressed to 2, the number of continuous shots increases to 20, and the decrease in sensitivity is reduced to 1/4.

(branching image sensor)
Non-Patent Document 2 discloses an image sensor 18 having a structure shown in FIG.

Two more gates 20 branch from each of the six multi-charge collection gates 19 . A sensor having such pixels is called a "branching image sensor".

FIG. 3 shows an example 22 in which the trajectory of an electron generated by an incident photon 21 is generated by Monte Carlo simulation. I decided to call this sensor "Fireworks" because of its inverted shape.

The number of continuous shots of fireworks is apparently 12 shots.

If two of the 12 are used as drains, the number of continuous shots will be 10. If you shoot with a macro pixel configuration of 2×2 pixels, you get 40 shots.

Development of a three-dimensional junction image sensor in which a plurality of semiconductor chips are joined is progressing. An example is shown in FIG.

Chips bonded to the sensor chip 126 include a driver chip 127 mounting a driver circuit, a memory chip 128 mounting a large number of memory elements, and the like.

Burst imaging with 10 to 40 shots is not the only way to shoot fireworks.

From the 12 floating diffusions 24 (see FIG. 3), if the signal charge packets that have reached each floating diffusion until then are sequentially sent to the memory chip 128 (see FIG. 10) that is joined, several hundred consecutive images can be produced at ultra high speed. You can shoot. This imaging method is called "quasi-continuous imaging".

In FIG. 3, a rectangular parallelepiped photodiode 23 with a square aperture having a smaller area than the pixel area is used instead of the p-well for collecting charges to the center of the pixel.

The purpose of Non-Patent Document 2 is to investigate the characteristics of the pixel structure on the surface side, so the structure of the photoelectric conversion layer is made as simple as possible.

If an on-chip microlens (not shown) with good performance is attached, the effective aperture ratio becomes sufficiently large.

Non-Patent Document 1 and Non-Patent Document 2 also disclose a pyramid charge collection structure (not shown), which is a high-performance charge collection structure that replaces the p-well.

The functions of burst imaging and quasi-continuous imaging with pyrotechnics are described.

Burst imaging of fireworks can capture substantially up to 40 images with a temporal resolution of 100 ps or less.

In FIG. 3, the charge storage area provided at the edge of each pixel consists of the floating diffusion 24 .

Using floating diffusion, the stored charge signal packets are converted into voltage signals and transferred to the outside with low noise. That is, in this sensor, the floating diffusion has two functions of accumulating/storing the signal charge and reading out the signal.

Human eyeballs make unconscious high-speed movements called saccades, and the maximum amplitude frequency is about 7 Hz. Images are generated one after another at different positions on the retina corresponding to the saccades, but the brain corrects them and recognizes them as images that do not shake.

Therefore, playback of continuous images at a faster rate of 10 frames/second appears to most people as smooth motion.

Television has adopted a playback speed of 25 Hz (PAL format, European system) or 30 Hz (NTSC format, Japanese and American system) with a little leeway.

Replaying 40 images taken with 2×2 macropixels of fireworks at 10 frames/sec will result in a continuous image of 4 seconds, which is the minimum required to activate the dynamic recognition function in human image recognition. time is secured. This makes it possible to see phenomena that are difficult to find from an array of still images.

Furthermore, since the image signal is stored as a floating diffusion, as will be explained below, if the noise level is allowed to be a little, quasi-continuous imaging of fireworks can be performed at a time resolution of 1 nanosecond, and continuous ultra-high-speed shooting of several hundred images is possible.

Continuous shooting does not require a drain during shooting, so the image signal saved in one floating diffusion at a certain time can be transferred while the 11 image signals are saved in the other 11 floating diffusions. .

That is, for the same imaging speed, the readout time from each floating diffusion is about 10 times longer than that of a normal image sensor having one floating diffusion per pixel.

The power consumption required to read out signals from each floating diffusion is proportional to the readout speed, and the readout noise is proportional to the square root of the readout speed.

Therefore, compared to the case where each pixel is provided with one floating diffusion, with the same power consumption, image signals can be continuously sent to the joined memory chips at about 10 times lower speed for the same power consumption. At this time, the read noise associated with the transfer to the memory chip is reduced to about 1/3.

At present, the world record for time resolution in signal readout using floating diffusion is about 10 nanoseconds as disclosed in Non-Patent Document 1. Therefore, if fireworks are used, several hundred consecutive images can be taken with a time resolution of nanoseconds, which is about 10 times faster than that.

(limit time resolution)
Non-Patent Documents 3 and 4 define the limit time resolution as shown in FIG.

Time resolution is not determined by the arrival time of signal electrons to the detection surface. Determined by the spread of arrival times. The average arrival time can be subtracted later.

Suppose that one signal electron group is instantaneously generated on a certain plane, and the next signal electron group is generated after a little delay. Diffusion and mixing extend the arrival time while traveling a certain distance to reach the detection surface. We will approximate this distribution with a normal distribution.

When the ordinate values of two normal distributions 25 and 26 with different mean values and equal standard deviations are added in FIG. 4, when the distance 29 between the two mean values 27 and 28 is twice the standard deviation, There is no dip at the peak 30 of the distribution and the two distributions are indistinguishable.

This condition was named non-dip condition. The time resolution for this condition is named theoretical limit time resolution.

At this time, the overlap between the two normal distributions is large. In image sensor terminology, this overlap is called crosstalk.

On the other hand, a distinct dip 32 appears when the distance 31 between the means is 3.3 times the standard deviation. In this case, the two distributions can be clearly separated.

At this time, the crosstalk becomes 5%, which is within a practically acceptable range. Therefore, we named this condition as the practical limit temporal resolution.
(Limiting factor for maximum shooting speed of burst image sensor)

The maximum imaging speed of the burst image sensor depends on the theoretical limit due to the physical properties of the photodiode and the operation speed of the circuit for imaging.

First, the factors limiting the critical time resolution in the photodiode will be explained. Horizontal and vertical motion of the signal electrons within the photodiode each constrains the time resolution.

A good example showing the effect of horizontal motion is shown in FIG. That is, the horizontal movement of electrons on the p-well when they gather at the center of the pixel greatly expands the time resolution.

If the photodiode width is infinitesimal, there is no horizontal motion. However, the positions of electrons generated by incident light within the photoelectric conversion layer follow an exponential distribution with the average generation depth as a parameter. The distribution of the generation depth causes vertical mixing of electrons arriving at the surface side at different times. Diffusion until it reaches the surface is superimposed on this.

Its theoretical minimum limit is the theoretical limit time resolution, which in the case of silicon photodiodes is 11.1 ps as derived in Non-Patent Document 3. In the germanium photodiode, it is 0.25 ps as disclosed in Non-Patent Document 4.

As shown in FIG. 4, when designed under the condition of the theoretical limit time resolution, the temporal overlap between consecutive electronic packets is large and the crosstalk between images is large.

As shown in Non-Patent Document 4, when crosstalk is suppressed to 5%, the practical time resolution of silicon photodiodes is about 45 ps, and that of germanium photodiodes is 1.45 ps.

The above values are obtained from the examination of the effect of vertical motion. When the width of the photodiode is reduced below a certain level, the effect of vertical motion becomes dominant within the photodiode over the effect of horizontal motion.

(Dependence of time resolution on drive voltage)
The operation speed of the circuit for photographing is divided into the photographing circuit mounted on the image sensor and the driving circuit placed outside the image sensor.

Furthermore, the imaging circuit on the image sensor depends on the moving speed of signal electrons in the circuit diffusion layer and the product RC of the capacitance C of each gate of the circuit diffusion layer and the electrical resistance of the wiring connected thereto.

As for the transistors of the driving circuit, those having a driving speed of about 50 GHz (time resolution of 20 ps) have been put into practical use.

By bonding the driver circuit chip to the sensor chip using three-dimensional bonding technology and providing one driver circuit for each pixel or macro pixel, the wiring distance can be reduced from several millimeters in the case of normal image sensors to 10 micrometers or less. As a result, the wiring resistance R becomes several hundredths. As a result, the drive circuit can drive with a time resolution of 10 ps.

(Time resolution of horizontal motion of signal charge)
There are roughly two reasons for the decrease in time resolution caused by the movement of signal electrons in the circuit diffusion layer. The mixing that occurs near the pixel center and the velocity lag and diffusion during transport of the signal charge to the floating diffusion that acts as a charge storage gate during imaging.

When the signal electrons are confined to the center of the pixel by firework, p-well, charge collection pyramid, etc., the signal electrons 22 bundle up and come down to the circuit diffusion layer on the surface side, as shown in FIG. Transfer this horizontally to one of the floating diffusions. The distance between the position far from the floating diffusion of the electron flux and the position close to it is currently about 2 micrometers.

These mix and reach the floating diffusion. This is one of the main causes of deterioration in temporal resolution due to horizontal mixing.

FIG. 5 shows the drift velocity 33 of intrinsic silicon and the drift velocity 34 of intrinsic germanium with respect to the electric field. Although there are a dotted line and a solid line, there is no significant difference due to the difference in crystal orientation.

The 95% saturation drift velocity of intrinsic silicon is about 9×10 ⁶ cm/sec for an electric field of 25 kV/cm as shown in FIG. Therefore, if the light is transferred horizontally at the saturation drift velocity, the time difference required to pass through about 2 microns is about 20 ps.

The standard deviation when these electrons mix is about 10 ps, and the practical time resolution due to horizontal mixing is 3.3 times that, which is about 30 ps.

That is, if an electric field close to the saturation electric field can be applied in the horizontal direction, the time resolution of the silicon image sensor will be about the same as the time resolution of 45 ps caused by the vertical mixing effect caused by the light penetration depth in the photodiode.

In practice, however, it is necessary to apply a voltage difference of at least 10 V or more between horizontally adjacent electrodes in order to transfer data in the horizontal direction with a saturated drift electric field. It can be said that it is practically impossible from the viewpoint of electric power.

FIG. 6 shows an example in which a practical horizontal potential gradient is examined. A potential distribution 36 along the dashed line 35 of FIG. 3 is shown. The downward direction in the figure is the positive potential.

High voltages of 1.7 V, 2.3 V and 3.5 V are applied to three electrodes 37, 38 and 39 shown in the figure in order. VL=0V is applied to all the electrodes 20 of the other outermost gates, and 0.9V is applied to the electrodes 19 of the other charge collection gates.

As a result, a potential difference of a little less than 2V is generated at a distance of about 7 microns before and after the floating diffusion on the upstream and downstream sides. That is, even though a large voltage difference of 3.5 V is applied to the electrodes, the average electric field is about 2.8 kV/cm, which is an order of magnitude smaller than the 95% saturation electric field of 25 kV/cm.

In addition, the potential along the transfer path 35 rises toward the floating diffusion 179 while undulating a little.

That is, the shape of the potential in the transfer direction becomes flat at the central portion of each electrode. The electric field at this time is very small, being slightly larger than zero.

On the other hand, the boundary between the electrodes draws a gradient instead of changing stepwise. That is, the electric field is much larger than that in the center of the electrode. The reason for forming the gradient is that the steep stepwise voltage change at the boundary between the electrodes cancels out and becomes smooth in the transfer path (the line connecting the bottom of the groove with the highest potential) several hundred nanometers above. It is from. This effect is called the electric field fringe effect.

As shown in FIG. 5, the smaller the electric field, the smaller the drift velocity. For this reason, the mixing effect and the diffusion effect are increased in the center of the pixel, and the diffusion effect is increased in the transfer path, so that the time distribution of arrival of electrons is greatly spread.

This problem can be alleviated by maximizing the minimum electric field in the path of the signal electrons. The solution is the linear potential 40 shown in FIG. 6, ie a constant electric field. Therefore, the voltages of the electrodes 37, 38, 39, etc. are set so that the potential distribution is as close to a straight line as possible.

Further downstream, as shown in FIG. 3, the signal charges are transferred from the large width gate 38 to the smaller width gate 39 . In such a case, an effect called the narrow channel effect causes opposite gradients in the transfer direction when the same voltage is applied to both gates. Therefore, it is necessary to apply a high voltage to the gate on the downstream side, but the narrow channel effect reduces the electric field compared to the case without the narrow channel effect.

(resistive gate)
In the 1980s when CCDs were put into practical use, resistive (high resistance) gate CCDs were used to prevent the electric field from becoming low at the center of the electrode. FIG. 7 shows a conceptual diagram. 2 and 3 are turned upside down.

In a normal CCD gate, a voltage is supplied by contacting the center of each gate electrode. The voltage across the electrodes is therefore constant across the board.

In FIG. 7, contacts 42 and 43 are placed across the gate 41 to give a small voltage difference of 0.5V. As a result, a voltage difference is generated in the electrode itself in the transfer direction, and the voltage difference is reflected in the upper transfer path, and the potential distribution 44 that is flat at the center of the electrode is improved to a potential distribution 45 that has a clear gradient even at the center.

Since the 1990s, process miniaturization has advanced, the electrode length has become shorter, and the electric field fringe effect at both ends of the electrode has spread to the center, making it unnecessary to use a resistive gate.

Even today, when a linear image sensor does not require a high-density process compared to a normal two-dimensional image sensor, a low-cost process is used to contact both ends of a relatively long electrode. In some cases, an electric field is applied to the electrodes.

In this case, a high-resistance electrode is used because the current constantly flowing through the electrode increases the power consumption. In the case of polysilicon electrodes, the impurity concentration is set to about 10 ¹⁷ /cm ³ or less. This is the origin of the name of the resistive gate CCD.

On the other hand, normal CCDs use low-resistance electrodes in order to keep the voltage constant across each electrode in as short a time as possible. At the time the CCD was invented, the electrodes were aluminum. At present, low resistance polysilicon with an impurity concentration of 10 ¹⁹ /cm ³ or more is used. In some cases, metals are added to increase speed.

T. G. Etoh, et al. , Light-in-Flight Imaging by a Silicon Image Sensor: Toward the Theoretical Highest Frame Rate, Sensors, 19(10), 2247, 2019. N. H. Ngo, et al. , A Pixel Design of a Branching Ultra-Highspeed Image Sensor, Sensors, 21(7), 25063, 2021. T. G. Etoh, et al. , The Theoretical Highest Frame Rate of Silicon Image Sensors, Sensors, 17(3), 483, 2017. N. H. Ngo, et al. , Toward the Super-Temporal-Resolution Image Sensor with a Germanium Photodiode for Visible Light-, Sensors, 21(7), 20(23), 6895, 2020.

The electric field along the transfer path of the signal charge in the circuit diffusion layer becomes flat at the center of the multiple electrodes in the path, causing a decrease in drift velocity and an increase in diffusion effect, and the synergistic effect greatly reduces the time resolution. .

The optimum is maximizing the minimum electric field. The solution is a linear potential (constant electric field).

Also, when transferring charge from a gate of one width to a gate of narrower width, a narrow channel effect occurs.

In addition, when a relatively wide electrode is used, due to random movement of the signal charge, the signal charge is not directed to the charge collection gate intended by design, but to another charge collection gate, thereby causing crosstalk between different signal charge packets. occur.

It consists of voltage generating devices for generating I (≧2) kinds of different voltages, wiring for transmitting the different voltages, and semiconductors arranged on a plane in M rows×N columns (M≧1, N≧1). A device (hereinafter referred to as a "sensor chip") having elements (hereinafter referred to as "pixels"), wherein a plane parallel to the plane is a C plane, a direction orthogonal to the plane is an A direction, and the A direction is When the opposite direction is called the B direction, each pixel consists of an electrical insulating layer (hereinafter referred to as "insulating layer") parallel to the C plane and a semiconductor layer (hereinafter "circuit diffusion layer") extending from the insulating layer in the A direction. ), a layer that generates charges by electromagnetic waves or charged particles (hereinafter referred to as a “photoelectric conversion layer”), and an electrode layer in which L (L≧1) electrodes are arranged in the B direction, substantially the circuit diffusion layer and the photoelectric conversion layer adjacent to each other, the surface facing the photoelectric conversion layer is substantially smaller than the surface facing the circuit diffusion layer; By providing two or more contacts that are electrically connected to wires that transmit different voltages, signal charges generated by light incident on the photoelectric conversion layer are transferred to the circuit diffusion layer and the photoelectric conversion layer. By sending to a predetermined circuit of the circuit diffusion layer through the surface of the surface on the photoelectric conversion layer side, the signal charge is prevented from straying into other circuits in the circuit diffusion layer and generating a false signal, In the circuit diffusion layer, signal charges can be sequentially transferred from the center of the pixel to the gates arranged in the radial direction, and a voltage gradient is generated in the voltage in the electrode in the line direction connecting the two or more contacts. As a result, an electric field larger than the minimum electric field in the absence of the voltage gradient is generated in the circuit diffusion layer in a direction parallel to the line connecting the two or more contacts, resulting in a signal charge transfer rate. To provide an imaging device with high time resolution by preventing the spread of the arrival time at the arrival point of signal charges.

Furthermore, one of the electrodes is provided with three or more contacts substantially equidistant from the substantial center of the pixel, so that signals are sent to three or more charge storage gates radially from the center with equal time resolution at high speed. It can transfer charge.

Furthermore, by providing two electrodes substantially equidistant from the substantial center of the pixel, the electrodes other than the electrode requiring ultra-high-speed signal transfer in the pixel center region are made into a single contact for the current to flow. can be eliminated to reduce power consumption.

Further, the impurity concentration of one of the electrodes is substantially 10 ¹⁸ /cm ³ or less, thereby reducing the current value generated in the electrode and providing an imaging apparatus with reduced power consumption.

a strip-shaped region in said circuit diffusion layer parallel to a line connecting two substantially adjacent contacts on one of said electrodes; impurity species in a region outside said strip-shaped region; Alternatively, the path of the signal charge can be limited to the belt-like region by varying the impurity concentration, and the random movement of the signal electrons causes the signal electrons to wander into the charge storage gates other than the predetermined charge storage gates. Crosstalk between images can be reduced.

Furthermore, since one of the electrodes is a concave polygon, it is possible to make a cut in a portion other than the electrode surface corresponding to the intended path of the signal charge in the circuit diffusion layer. It is possible to prevent generation of excessive current, reduce power consumption, and reduce crosstalk.

A photograph showing flight of light captured by an image sensor with a time resolution of 10 nanoseconds. FIG. 1 shows a back-illuminated multi-charge collection gate image sensor with 1 ns time resolution and functionality. Branching image sensor: Cross-sectional view of fireworks and diagram showing electrode structure. The figure which shows the definition of the limit time resolution in case arrival time distribution is normal distribution. FIG. 4 shows the drift velocity of silicon and germanium versus electric field. FIG. 4 is a diagram showing potential shapes of horizontal transfer paths of fireworks; FIG. 4 is a diagram showing a comparison of potential distribution shapes within circuit diffusion layers of a resistive gate and a normal gate; The figure which shows a camera. Fig. 3 shows means for generating different voltages; Outline drawing of an image sensor. FIG. 4 is a diagram showing the outline of the surface of an image sensor chip; BRIEF DESCRIPTION OF THE DRAWINGS Sectional drawing of the 1st Embodiment of this invention. FIG. 2 is a diagram showing a schematic shape of an electrode according to the first embodiment of the present invention; FIG. 4 is a diagram showing operation timings of drive voltages according to the first embodiment of the present invention; FIG. 4 is a diagram showing an example of fireworks, electrodes and potential distributions according to the first embodiment of the present invention; FIG. 5 is a diagram showing a comparison of potential distribution between fireworks and the first embodiment of the present invention; FIG. 7 is a diagram showing impurity doping shapes of electrodes and circuit diffusion layers according to the second embodiment of the present invention; FIG. 10 is a diagram showing impurity doping shapes of electrodes and circuit diffusion layers according to a third embodiment of the present invention;

(First embodiment)
(Structure of the first embodiment)
[Shooting device]
FIG. 8 shows a configuration diagram 100 of an imaging apparatus according to the first embodiment of the present invention. Incident light 101 enters an image sensor 106 via an optical system, ie filter 102 , lens 103 , diaphragm 104 and shutter 105 .

An image sensor is attached to the tip of the camera unit 107 . The image sensor is controlled by an imaging control circuit 108 and a signal readout circuit 109 . A buffer memory 110 for temporarily storing the digital image signal is connected to the signal reading circuit. The buffer memory is connected to the communication circuit 111 and further to the image processing device 112 in the control computer. The control computer also incorporates a control circuit 113 and an internal recording device 114 for controlling the entire system. Also connected to the control computer are an external recording device 115, a display 116, a console 117, a mouse 118, a trigger device 119 for starting and stopping photography in synchronization with the occurrence of a phenomenon to be photographed, and an illumination device 120. .

FIG. 9 shows the imaging control circuit 108. As shown in FIG. A signal waveform generated by the FPGA 121 and a constant voltage generated by the constant voltage generation device 122 are synthesized to generate a dynamic voltage waveform by the voltage generation device 123 , which is output from the wiring 124 .

Control is performed according to the program incorporated in FPGA121. The FPGA 121 also controls the signal readout circuit.

[Image sensor]
FIG. 10 shows the outline of the image sensor 125. As shown in FIG.

Three IC chips are bonded to form one image sensor 125 . A chip on the light receiving side is called a sensor chip 126 . A drive circuit chip 127 and a memory chip 128 are bonded to the sensor chip.

The sensor chip 126 and the drive circuit chip 127 are electrically connected by chip-to-chip contacts (not shown). Also, the memory chip 128 is electrically connected to the drive circuit chip 127 via through electrodes (not shown) and chip-to-chip contacts (not shown).

A plurality of voltage waveforms sent from the voltage generating device 123 are transmitted through wiring and circuits (not shown) on the driving circuit chip 127, and the time difference between the transfer times to each pixel is substantially negligible. , and sent to the electrode contacts of each pixel in the sensor chip 126 .

[Sensor chip]
A rear surface voltage VB is supplied to the rear surface 129 of the sensor chip 126 from a pad 130 through a metal wiring 131 .

FIG. 11 shows a plan view 132 of sensor chip 126 .

At the center of the sensor chip 126 is a light receiving surface 133 consisting of 1024×1024 pixels. There is a pad area 134 and a multiplexer area 135 around it.

The size of one pixel is 11.4 microns by 11.4 microns. Therefore, the light receiving surface is 11.6736 mm×11.6736 mm. The chip size is 20 mm×20 mm.

FIG. 12 shows a cross-sectional view 136 of one pixel of the sensor chip 126 of the first embodiment. In the figure, the top is the back side.

Each pixel has a condensing means consisting of an on-chip microlens on the back side, but it is not shown because it is self-explanatory.

The back surface 129 of the sensor chip 126 is covered with a light shielding layer 137 and has an opening 138 .

Incident light 139 is focused substantially onto the aperture by a focusing means (not shown) such as an on-chip microlens.

A photoelectric conversion layer 140 and a circuit diffusion layer 141 made of a silicon layer are provided under the light shielding layer 137, and a silicon oxide film 142 electrically separates the light shielding layer 137 and its opening from the silicon layer.

The lower surface (surface side) of the circuit diffusion layer 141 is an insulating layer 143 made of silicon oxide, and the electrode layer 144 is joined to the lower surface.

As shown in FIG. 10, a drive circuit chip 127 is further bonded to the lower surface (not shown in FIG. 12).

The pixel size is 11.4 microns, but the width of the opening of the light shielding layer 137 and the width of the photoelectric conversion layer 140 are both 5 microns and square.

The photoelectric conversion layer 140 has a thickness of 11.1 microns, and the circuit diffusion layer 141 has a thickness of 1 micron. The underlying insulating layer 143 is a silicon dioxide film with a thickness of 20 nanometers. The electrodes are made of polysilicon and are 0.02 microns thick.

The photoelectric conversion layer 140 and the circuit diffusion layer 141 are doped with phosphorus ions at a concentration of 10 ¹⁴ /cm ³ before impurity implantation. The semiconducting properties of silicon at this concentration are substantially the same as those of intrinsic silicon.

The side surface of the photoelectric conversion layer 140 is covered with an insulating layer 145 of silicon dioxide having a thickness of 20 nm, and the space 146 outside thereof is filled with polysilicon.

The back surface side (upper side in the drawing) and side surfaces of the photoelectric conversion layer 140 are doped layers 147 of boron ions having a thickness of 0.1 μm and a concentration of 10 ¹⁸ /cm ³ .

On the surface side of the circuit diffusion layer, a dodecagonal transfer path layer 148 with a thickness of 0.3 μm centering on the center of the pixel and entirely doped with phosphorus ions at a concentration of 10 ¹⁷ /cm ³ is formed.

Under the insulating layer, one dodecagonal electrode 149 is arranged around the center of the pixel.

FIG. 13 shows the arrangement of one dodecagonal electrode 149, twelve contacts 150, and a readout circuit 151 of the 12 on the surface side.

Each readout circuit 151 consists of a floating diffusion 152 , a reset gate 153 and a reset drain 154 .

(Functions and operations of the first embodiment)
A voltage of VB=-20V is applied to the rear surface, and a voltage of 3V is applied to the reset drain.

A high voltage VCH=3.5V and a low voltage VCL=0V are alternately applied to contact 150 .

FIG. 14 shows the drive voltage pattern while taking 10 images in the case of burst imaging. In this case, a voltage waveform 155 is sent to contacts 157 and 159 . That is, the voltage CVH is sent before photographing, and the voltage CVL is sent during photographing. CVL is sent to other contacts before and after shooting.

As a result, electrons generated in the photoelectric conversion layer before and after photographing are discharged from the floating diffusion 160 and the floating diffusion 161 through the reset drain 162 and the reset drain 163 to the outside of the pixel.

CVH is sent to the other contacts in sequence, and the signal charge is stored in each floating diffusion in sequence. The operation of the reset gate during this period is well known and is not shown.

Therefore, the number of shots in burst imaging is ten.

On the other hand, in quasi-continuous readout to the memory chip (driving pulses are not shown), the floating diffusions 160 and 161 perform the same readout operation as the other floating diffusions.

The signal acquired by one floating diffusion is slowly transferred to the memory chip while the signals are acquired sequentially by the other 11 floating diffusions.

FIG. 15 shows the potential distribution 158 of the circuit diffusion layer when CVH is applied to the electrode arrangement 156 of the fireworks, the electrode 149 of the first embodiment, and the contact 157, and CVL is applied to the other electrodes.

Although the fireworks are divided into 19 gates, the size of the outline of all of them is the same as that of the single gate 149 in the first embodiment. Otherwise, the sizes of the components of the first embodiment are the same as the sizes of the components of the corresponding elements of the fireworks.

The voltage difference (VCH-VCL) applied to the twelve contacts 150 arranged on the outer edge of the first embodiment is 3.5 V, and the voltage difference of the voltage applied to the twelve gates on the outer edge of the fireworks is also 3.5V.

A smooth and steep potential gradient is generated from the contact 159 to the contact 157 .

FIG. 16 shows potential distribution 164 along dashed line 35 in FIG. 3 and potential distribution 166 for a 3.6 micron path from contact 159 to contact 157 along dashed line 165 in FIG. The path length of potential distribution 164 is also matched to the path length of potential distribution 166 . Since the potential difference is about 1 V (=0.6 V-(-0.4 V)), dividing by 3.6 microns gives an average electric field of about 2.6 kV/cm, which is 1 of the 95% saturation electric field of 25 kV/cm. /10.

As shown in FIG. 16, the electric field in the flat portion of the potential distribution 164 is very small, but the potential distribution 166 has a substantially ideal distribution in which the potential rises smoothly.

The diameter of the bundle 22 of signal electrons descending from the photoelectric conversion layer shown in FIG. spreads.

In order to exclude this effect and evaluate the spread of time resolution due to delay and diffusion in the transfer path, a large number of electron trajectories starting from the pixel center 10 were generated by the Monte Carlo method for the potential profile of FIG. , the standard deviation of the arrival time to the contact 157 was obtained.

The standard deviation was determined for the first embodiment and fireworks. It was 35.2 ps in the first embodiment, and 75.2 ps in fireworks.

A practical temporal resolution is 3.3 times the standard deviation.

Therefore, the practical limit time resolution is 116.2 ps for the first embodiment and 248.2 ps for fireworks. As already mentioned, the practical limit time resolution caused by vertical mixing in the photodiode is about 45 ps, so it can be seen that the diffusion in the horizontal transfer is the dominant factor of the time resolution in fireworks.

From the above, it can be seen that the time resolution is reduced to 1/2 or less by the first embodiment of the present invention.

The first embodiment includes a voltage generating device (123) for generating I (≧2) kinds of different voltages, wiring (124) for sending the different voltages, and M rows×N arranged on a plane. and a sensor chip that is a device comprising pixels that are elements made of semiconductors in rows (M≧1, N≧1), wherein the plane parallel to the plane is the C plane, and the direction perpendicular to the plane is the A direction. , when the opposite direction of the A direction is called the B direction, each pixel includes an electrically insulating layer (143) parallel to the C plane and a circuit diffusion layer which is a semiconductor layer in the A direction from the insulating layer (143). (141), a photoelectric conversion layer (140) which is a layer that generates charges by electromagnetic waves or charged particles, and an electrode layer (144) in which L (L≧1) electrodes (149) are arranged in the B direction. and the circuit diffusion layer (141) and the photoelectric conversion layer (140) are adjacent to each other, and the photoelectric conversion layer (140) side faces the circuit diffusion layer (141) and one of said electrodes (149) comprises two or more contacts (150) electrically joining wires (124) carrying said different voltages. .

(Second embodiment)
FIG. 17 is a diagram showing an arrangement diagram 168 of electrodes 167 and contacts according to the second embodiment of the present invention and an ion doping pattern 169 on the surface side of the circuit diffusion layer superimposed.

In the second embodiment, in addition to the 12 contacts on the outer edge, the contact at the center of the pixel and the 6 contacts at the midpoint between the contact on the outer edge and the contact at the center of the pixel are combined to form a total of 19 contacts. are provided.

In the second embodiment, the distance between the electrodes is about 1/4 of that in the first embodiment. is increased, the total power consumption is reduced by an order of magnitude.

In the second embodiment, along the line connecting the contacts in the direction of branching, i.e. along the path of the signal electrons, a phosphorus ion doping 169 with a thickness of 0.1 microns, a width of 0.3 microns and a concentration of 10 ¹⁷ /cm ³ is provided. ing.

As a result, the narrow channel effect due to the sudden narrowing of the width of the transfer path can be effectively suppressed, and the speed can be further increased.

In addition, since the potential groove is formed, crosstalk caused by signal electrons wandering into a floating diffusion different from the floating diffusion that collects charges by horizontal random movement of signal electrons is reduced.

(Third Embodiment)
FIG. 18 shows an ion doping pattern 171 on the surface side of the circuit diffusion layer according to the third embodiment of the present invention, an arrangement diagram 172 of electrodes and contacts, and a diagram 173 showing both of them superimposed.

There are three types of ion concentration doping. Doping 174 along the electrode geometry is 10 ¹⁶ /cm ³ phosphorus ion doping. The doping 175 along the branched channel is 10 ¹⁷ /cm ³ of arsenic ion doping. Floating diffusion and reset drain contact doping 176 is arsenic ion doping of 10 ¹⁹ /cm ³ .

A feature of the transfer electrode 177 is that it is branched. In the case of a convex dodecagonal electrode, when the voltage applied to the contact changes, a locally eddy current is generated. This phenomenon is suppressed in the third embodiment.

Further, a reset gate electrode 178 is provided.

(Other embodiments)
Embodiments are not limited to the above examples. For example, the branched end electrodes in FIG. 18 may be separated by a narrow gap and each provided with one contact. once. Electrons that have reached the branched tip have a very low probability of returning to the original direction, so they may be sent to the flowing diffusion at a relatively low speed. This saves power consumption.

Also, the signal charges may be holes. In this case the doping polarity of the semiconductor is reversed.

The photoelectric conversion part may be of an inverted pyramid type, and if the pixel size becomes smaller in the future, the potential separation may be by p-well.

The present invention is summarized as follows.

An object of the present invention is to provide an image sensor and a photographing apparatus capable of continuously taking 10 to 40 images at a photographing speed equivalent to 100 billion images/second, and 100 images continuously at a photographing speed equivalent to 1 billion images/second. Therefore, it is necessary to control the retardation of the horizontal movement of signal electrons in the pixels of the image sensor to make the photographing speed very high.

The solution to the problem is as follows. A back-illuminated image sensor consists of an upper photodiode, a lower circuit diffusion layer, and an insulating layer that isolates them except for the center of the pixel. , and sequentially transferred radially from the center of the pixel in the circuit diffusion layer and stored. Electrodes for transfer have multiple contacts supplied with different voltages. As a result, a radial voltage gradient is generated in the electrode, which increases the minimum electric field in the transfer path of signal electrons in the circuit diffusion layer, suppresses the spread of transfer time, and enables imaging with higher time resolution. to

For example, in the pump-probe method, electrons are excited by irradiating an ultra-short pulse laser, and the relaxation process is illuminated for a short time with a probe laser and photographed with an ordinary camera. Observing the dynamic changes of ultrafast phenomena by repeating photography while shifting the time.

A phenomenon with a temporal resolution of about 100 ps can be photographed only once by using a camera equipped with the image sensor of the present invention. In addition, continuous shooting of phenomena with poor reproducibility or phenomena that require a lot of preparation can be done in one shot.

In addition to such direct imaging applications, there are many applications as sensors for scientific measurement technology such as LIDAR, FLIM (Fluorescence Lifetime Imaging), and Imaging TOF MS.

1 Sixth image of flying light captured by an image sensor with a temporal resolution of 10 nanoseconds 2 An image of flying light captured by a normal video camera 3 Cross-sectional view of one pixel of an image sensor with a temporal resolution of 10 nanoseconds 4 Layout of electrodes on the front side (lower side in the figure) 5 Back side (upper side in the figure)
6 Light incident on the back surface 7 Photoelectric conversion layer (photodiode)
8 Example of reproducing the trajectory of generated electrons by Monte Carlo simulation 9 Time (horizontal axis) for the electron to reach a certain depth (vertical axis) from the position where the electron was generated
10 pixel center 11 6 charge collection gates radially arranged from the pixel center 12 6 charge collection gates radially arranged from the pixel center 13 p-well
14 Circuit Diffusion Layer on Front Side 15 Charge Storage Gate 16 Readout CCD Gate 17 Barrier Gate 18 Outline of Branching Image Sensor (Fireworks) 19 6 Multiple Charge Collection Gates 20 2 Branched to 6 Multiple Charge Collection Gates gate 21 incident photon 22 electron trajectory generated by Monte Carlo simulation 23 rectangular photodiode with square aperture 24 floating diffusion 25, 26 two normal distributions with different mean values and equal standard deviations 27, 28 mean value 29 mean Distance between values 30 Peak of the distribution of the sum of two normal distributions with different means and equal standard deviations 31 Distance between means 32 Density of the distribution of the sum of two normal distributions with different means and equal standard deviations 33 Relationship between electric field and drift velocity of intrinsic silicon 34 Relationship between electric field and drift velocity of intrinsic germanium 35 Dotted line in FIG. Contact of the stive gate 44 Potential profile with a flat portion when there is one contact 45 Potential profile with a clear gradient even at the center of the electrode when there are two contacts 100 Imaging device according to the first embodiment of the present invention 101 Incident light 102 Filter 103 Lens 104 Diaphragm 105 Shutter 106 Image sensor 107 Camera part 108 Shooting control circuit 109 Signal readout circuit 110 Buffer memory 111 Communication circuit 112 Image processing device in control computer 113 Control Circuit 114 Internal Recording Device 115 External Recording Device 116 Display 117 Console 118 Mouse 119 Trigger Device for Starting and Stopping Shooting 120 Lighting Device 121 FPGA
122 constant voltage generation device 123 voltage generation device 124 wiring 125 outline of image sensor 126 sensor chip 127 drive circuit chip 128 memory chip 129 rear surface of sensor chip 130 pad 131 metal wiring 132 plan view of sensor chip 133 light receiving surface of sensor chip 134 pad Region 135 Multiplexer region 136 Cross-sectional view of one pixel of the sensor chip of the first embodiment 137 Light shielding layer 138 Opening 139 Incident light 140 Photoelectric conversion layer 141 Circuit diffusion layer 142 Silicon oxide film 143 Insulating layer made of silicon oxide 144 Electrode Layer 145 Silicon dioxide insulating layer on the side surface of the photoelectric conversion layer 146 Space outside the photoelectric conversion layer 147 Boron ion doped layer on the back and side surfaces of the photoelectric conversion layer 148 Centered on the pixel center on the front surface side of the circuit diffusion layer Dodecagonal phosphorus ion doped layer 149 One dodecagonal electrode centered on the pixel center under the insulating layer 150 12 contacts 151, 12 readout circuits 152 floating diffusion 153 reset gate 154 reset drain 155 as drain Voltage waveform sent to the electrode of the gate used 156 Electrode arrangement of fireworks 157 Contacts that sent CVH 158 Potential distribution of the circuit diffusion layer when sending CVH to one contact and CVL to the other contact 159 Contact 160 Floating diffusion 161 Floating diffusion 162 Reset drain 163 Reset drain 164 Potential distribution along chain line 35 in FIG. 3 165 Chain line in FIG. 13 166 Potential distribution from contact 159 to contact 157 along chain line 165 in FIG. Contact of the electrode of the second embodiment 169 Ion doping pattern of the same electrode 170 Diagram showing the layout of the contact of the same electrode and the ion doping pattern superimposed 171 Ion doping on the surface side of the circuit diffusion layer of the third embodiment Pattern 172 Layout of same electrode and contact 173 Overlapping of both 174 Phosphorus doping shape along electrode shape 175 Arsenic ion doping shape along branched channel 176 Contact doping of floating diffusion and reset drain ring 177, 179 floating diffusion 178, 180 reset drain 181 inhibition layer

Claims (6)

a voltage generating device (123) that generates I (≧2) different voltages;
a wire (124) carrying the different voltages;
and a sensor chip, which is a device comprising pixels that are elements made of semiconductors arranged on a plane in M rows×N columns (M≧1, N≧1),
When the plane parallel to the plane is called the C plane, the direction perpendicular to the plane is called the A direction, and the opposite direction of the A direction is called the B direction,
Each pixel is
An electrically insulating layer (143) parallel to the C plane, a circuit diffusion layer (141) that is a semiconductor layer in the A direction from the insulating layer (143), and a photoelectric layer that generates charges by electromagnetic waves or charged particles. A conversion layer (140) and an electrode layer (144) in which L (L≧1) electrodes (149) are arranged in the B direction, substantially laminated,
a surface of the adjacent surfaces of the circuit diffusion layer (141) and the photoelectric conversion layer (140) facing the photoelectric conversion layer (140) is substantially smaller than the surface facing the circuit diffusion layer (141);
An imaging device, wherein one of said electrodes (149) comprises two or more contacts (150) electrically joining wires (124) carrying said different voltages. The imaging device of claim 1,
An imager, wherein one of said electrodes (149) comprises three or more contacts (157) substantially equidistant from a substantial pixel center. The imaging device according to claim 1 or claim 2,
An imaging device comprising two electrodes (149) substantially equidistant from a substantial pixel center. The imaging device according to any one of claims 1 to 3,
An imaging device, wherein the impurity concentration of one of the electrodes (149) is substantially 10 ¹⁸ /cm ³ or less. The imaging device according to any one of claims 1 to 4,
a strip in said circuit diffusion layer (141) parallel to a line joining two substantially adjacent contacts (157) on one of said electrodes (149); Imaging device in which the impurity species or impurity concentration in the external region is different. The imaging device according to any one of claims 1 to 5,
An imaging device, wherein one of said electrodes (149) comprises a concave polygon.

Images